Learning Robust Manipulation Skills with Guided Policy Search via Generative Motor Reflexes

Abstract

Guided Policy Search enables robots to learn control policies for complex manipulation tasks efficiently. Therein, the control policies are represented as high-dimensional neural networks which derive robot actions based on states. However, due to the small number of real-world trajectory samples in Guided Policy Search, the resulting neural networks are only robust in the neighbourhood of the trajectory distribution explored by real-world interactions. In this paper, we present a new policy representation called Generative Motor Reflexes, which is able to generate robust actions over a broader state space compared to previous methods. In contrast to prior state-action policies, Generative Motor Reflexes map states to parameters for a state-dependent motor reflex, which is then used to derive actions. Robustness is achieved by generating similar motor reflexes for many states. We evaluate the presented method in simulated and real-world manipulation tasks, including contact-rich peg-in-hole tasks. Using these evaluation tasks, we show that policies represented as Generative Motor Reflexes lead to robust manipulation skills also outside the explored trajectory distribution with less training needs compared to previous methods.1 Philipp Ennen, Pia Bresenitz, Rene Vossen, Frank Hees are with the

Multi-Contact Task and Motion Planning Guided by Video Demonstration

“…Learning-based methods. Policy search methods, for example, reinforcement learning, are widely used to solve manipulation tasks [22], [23], [24]. However, these methods usually require substantial training time, and once the policy is learned, it usually does not generalize well to different scenes, for example, if the furniture is moved.…”

Goal-Conditioned Reinforcement Learning within a Human-Robot Disassembly Environment

“…Currently, the application of RL in assembly is focused on three lines of research: improving performance, sample efficiency, and generalization capability and narrowing the simulation-reality gap. Performance improvement spans different domains although research generally focuses on accuracy [15], safety [16], robustness [17], and contact stability [18]. Other studies, such as [19,20], analyzed stability from a different perspective considering that any state trajectory must be bounded and tend to the target position required by the task.…”